MGSC 291 Whitcomb
cognitive bias
1. anchoring effect 2. availability bias 3. confirmation bias 4. overconfidence bias
solutions to anchoring effect
1. ask for multiple, independent opinions 2. reconstruct multiple answers from scratch 3. downplay initial information; work from fundamentals 4. use experts
to be regarded as a candidate for normality, a random variable should
1. be measured on a continuous scale 2. possess a clear center 3. have only one peak (unimodal) 4. exhibit tapering tails 5. be symmetric about the mean (equal tails) most physical measurements in the real world would resemble normal distributions
basic roadmap to designing effective graphics (3 things)
1. determine your message 2. select the most effective format to display your message 3. design the data to show the most important parts
types of statistical lies
1. incorrect data 2. ignoring the baseline 3. arbitrary comparisons 4. misleading comparisons
design the data to show the most important parts
1. make the data clear and straightforward 2. remove all components that aren't necessary 3. highlight the most important parts; mute supporting elements 4. encourage the eye to compare data
normal Probability Distribution
1. normal or gaussian (or bell shaped) distribution was named for Karl Gauss 2. defined by two parameters µ and σ 3. denoted N(µ ,σ) 4. domain is -infinity < X < +infinity (continuous scale) 5. 99.7% of the area under the normal curve is included in the range between three standard deviations of the mean 6. the distribution is symmetric, bell shaped
sampling distributions for proportions
1. sample given as a percent 2. qualifies for normality Then the resulting distribution for the sample is normal and has the following characteristics... p is approximately normal with formulas on formula sheet
Tufts's rules for splaying quantitative information
1. show your data 2. use graphics 3. avoid chart junk 4. utilize data ink 5. use labels 6. use micro/macro 7. separate layers 8. use multiples 9. use color 10. understand narrative
Recognizing Poisson process
1.) an event of interest occurs randomly over time (or space) 2.) the average arrival rate (λ) remains constant 3.)the arrivals are independent of each other 4.) the random variable (X) is the number of events within an observed time interval
all discrete probability distributions must satisfy
1.) probabilities being between or equal to 0 and 1 2.) sum of possibilities of all values of X must sum to 1 probabilities of X can be anything
expected value
E(X) μ: mean--weighted average
special law of multiplication: if events A and B are independent then
P(A ∩ B) = P(A)P(B)
calculating conditional probabilities
P(A ∩ B) = P(A|B) P(B) P(A|B) = P(A ∩ B) / P(B)
if A and B are mutually exclusive, then the general law of addition can be simplified to...
P(A)+P(B)
When events A & B are mutually exclusive,
P(A|B) = 0 and P(B|A) = 0. cannot both be true --> single coin toss
When events A & B are independent
P(A|B) = P(A) and P(B|A) = P(B)
conditional probability
P(A|B): "the probability of A given B" P(A ∩ B) / P(B) for P(B)>0
deceptive graphs
Page 92 error 1: nonzero origin error 2: elastic graph proportions (stretching the Y or X axis) error 3: dramatic titles and distracting pictures error 4: 3-D and Novelty graphs--distort the bar volume error 5: chart junk error 6: data to ink ratio error 7: area trick: bars of graph misstates the true proportions
complement rule
The probability of the complement of an event, P(Ac), is equal to one minus the probability of the event.
standardized dara (z score)
a general approach to identifying unusual observations is to redefine each observation in terms of its distance from the mean in standard deviations
a probability distribution for a random variable X is
a listing of all possible values of X and their probabilities of occurring
Bernoulli experiments
a random experiment that has only two accounts to create a random variable we call one outcome a success X=1 and one outcome a failure X=0 the probability of success is denoted 𝜋 and probability of failure is 1-𝜋 ^both sum to 1
sample data set
a subset of the population characteristics are called statistics in most cases we cannot study all the members of a population
example of dependent event
age and phone use / arthritis ... knowing a persons age would effect the probability that the individual uses test messages or has arthritis may be casually related but does not prove cause and effect --- only means that knowing event B has occurred will effect the probability that A will occur
Probabilities as areas-continuous
areas under curves P(a<X<b) is the integral of the probability density function over the interval a to b because P(X=a)=0 entire area under PDF must be 1
binomial distribution
arises when a Bernoulli experiment is related n times each Bernoulli trial is independent so that the probability of success 𝜋 remains constant on each trial
why do we call the poisson distribution the model of arrivals (customers, defects, accidents)
arrivals can be regarded as Poisson events if each event is independent ex) X= # of customers arriving at a bank ATM in one minute
Bar and Column charts
column chart is a vertical bar chart numerical values on y axis category labels on x axis the height or length of each bar
anchoring effect
common human tendency to rely too heavily on the first piece of information offered when making decisions ex) originally for x amount but well give it to you for x-100!!! feel like you got a great deal
sampling is the basis for
confidence intervals hypotheis testing
contingency table
cross-tabulation of frequencies into rows and columns the intersection of each row and column is a cell that shows frequency
when P(A) differs from P(A|B) the events are ...
dependent
exponential distribution
describes the distribution of time between two events when the count of events has a poisson distribution (λ = mean # events per time)
poisson probability distribution
describes the number of occurrences within a randomly chosen unit of time (minute or hour) or space (square foot, linear mile) events must occur randomly and independently over a continuum time or space represented by dots
random variables can be either
discrete or continuous
how to calculate inverse probabilities for uniform distribution
ex) it is equally likely to arrive at any time within 20 minutes after he sets foot on the platform. What is the length of time he would wait such that there is a 30% chance he will wait this amount of time or longer? Area = .30= Length x height = (20-x0)×.05 .30=(20−_𝑥0).05 6.0=20−x0 𝒙𝟎=𝟏𝟒.𝟎
rules of probability
expressed as decimals and should add up to 1.0
histogram
graphical representation of a frequency distribution no gaps b/t bars
discrete random variable
has a countable number of distinct values number of sixes in 4 dice rolls 0,1,2,3,4
sampling when mean is normal
has to have a large number of trials need more than 30 observations in a sample underlying distribution must be normal
why is the poisson distribution called the model of rare events
if the mean is large, we can reformulate the time units to yield a smaller mean λ=90 events per hour = λ=1.5 events oer minute
collectively exhaustive
if their union is the entire sample space at least one of the events has to occur
tables
individual values will be looked up/compared, precise values are required
joint probabilities (CT)
intersection of 2 events (each cell in a contingency table)
random variables
is a function or rule that assigns a numerical value to each outcome in the sample space of a random experiment use X when referring to random variable, while specific variables of x are shown in lower case
probability density function
is used ti describe continuous random variables
probability mass distribution
is used to describe random discrete variables
skewed right
long tail of histogram points right (most data on left) mean>median
a cumulative distribution function
may be used to describe both discrete and continuous random variables
graphs
message is the shape of the values
skewed left
negative skewness long tail of histogram points left (mostly data on the right) mean<meadian
in a binomial experiment we are interested in the
number of success in n trials, so the binomial random variable X is the sum of n independent Bernoulli random variables X=X1+X2+X3+X4 +Xn
confirmation bias
occurs when decision makers seek out evidence that confirms their previously held beliefs while discounting or demising evidence that supports differing conclusions
proportions can always be expressed as a percentage buttt
percentages cannot always be expressed as proportions
random sampling allows us to draw valid conclusions about
populations-all the people, items, or objects of interest- from random samples drawn from those populations
addition rule
probability that event A or B occurs (at least one of the events happen P(A U B) = P(A) + P(B) - P(A ∩ B) union--either or both events will occur
confidence examples over just right under
provide a low and high guess of 10 items..90% sure that the correct answer falls between the two. if you successfully meet the challenge you should have 10% misses-one miss overconfidence--two or more of my intervals did not contain correct answer under confidence-- all intervals contained right answer
Bayes' theorem
provides a method of revising probabilities to reflect new information the prior (marginal) probability of an event B is revised after event A has been considered to yield a posterior (conditional) probability
continuous random variable
random variable that arises from measuring something infinite number of outcomes don't have a probability attached to each outcome, we have a density f(x) f(x)>_0 ex) waiting time until a customer arrives can have decimal values
Poisson distribution events occur...
randomly and independently over a continuum of time or space
conditional probability (CT)
restricting ourselves to a single row or column ex. salary gains are small given that MBA tuition is large
Continuous expected value and variance
same as discrete except instead of ∑ , it uses the integral sign ∫. Integrals are taken over all X-values
standard deviation
single number that helps us understand how individual values in a data set vary from the mean
shape of binomial distribution
skewed right when 𝜋 < .5, symmetric when 𝜋=.5 and skewed left when 𝜋>.5
Uniform Continous distribution
sometimes noted as U(a, b) Since PDF is rectangular you can easily verify that the area under the curve is 1 by multiplying the base (b-a) by its height 1/ (b-a) f(x) the density is constant
mean
sum of all data values divided by the number of data items
variance of a discrete random variable
sum of the squared deviations about its expected value, weighted by the probability of each X-value standard deviation squared
sample variance
s² replace μ with x bar
symmetric
tails of histogram are balanced mean=meadian
median
the 50th percentile
availability bias
the belief that if something can be recalled, it must be important --people tend to weight their judgements toward more recent information --it is easier to recall consequences if those consequences are bigger ex) if asked if more ny students go to usc than tennesse students it would be based on personal examples ex) if people understood the odds of winning the lottery no one would ever buy a ticker. However since the jackpot winners are advertised so frequently, people forget about the vast majority of people who haven't won a cent
population data set
the complete set of individuals characteristics are called parameters
intersection
the event consisting of all outcome in the sample space that are contained in both event A or B joint probability is denoted as P(A ∩ B)
mode
the most frequently occurring data value
as λ increases
the poisson becomes less skewed on the graph
marginal probability
the relative frequency that is found by diving a row or column by the total sample size
expected value for the mean of a sample size n is
the same as expected value for an individual observation long run average
standard error
the standard deviation of the sampling distribution the dispersion of a sample will be lower than that for a single observation each sample could contain highs and lows which cancel out
the probability of 2 independent events occurring simultaneously is the product of
their separate probabilities P(A1 ∩ A2 ∩ A3)= P(A1)P(A2)P(A3)
how to counter the availability bias
try to think of instance of the event that aren't so memorable ex) all instance someone swam at beach and wasn't attacked by shark ex) all the people who weren't murdered in the dt area
coefficient of variation
unit free measure of dispersion the standard deviation expressed as a percent of the mean when SD exceeds the mean the CV can exceed 100% want a lower number useful in comparing variables measured in different units
using the standard normal table to calculate normal probabilities
us it to look up the probability that X takes on a value in a given range, or given interval. We can use the standard normal probability table for any normal random variable (no matter what its mean and standard deviation) by transforming the normal random variable X into a standard normal random variable Z using the following formula Z=X-𝜇 / 𝜎
scatterplot
used to depict two potentially related variables -each point is a pairing -linear, curvilinear -positive vs negative relationships
line graph
used to display a time series, to spot trends or to compare time periods no vertical grid lines only horizontal numerical variable on y axis and time units on x axis use zero origin numerical labels omitted data markers
overconfidence effect
well established bias in which a persons subjective confidence in her or his judgements is reliably greater than the objective accuracy of those judgements, especially when confidence is relatively high
independent
when knowing that event B has occurred does not affect the probability that event A will occur ex) event A is independent of event B if the conditional probability P(A|B) is the same as the unconditional probability P(A)--> if the probability of event A is the same whether event B occurs or not
when do we treat discrete variables as continuous?
when the range is large ex) exam scores-- range from 0-100 but are often treated as continous ex) number of people in Richland County who have purchased flood insurance policy
problem with confirmation bias
you selectively filter what info to choose to pay attention to and value ex) person believes left handed people are more creative than right handed people, now every time they meet a left handed person that is also creative they place even greater support on the "evidence" --could be scientifically proven by experts, but also discounting examples that do not support the idea
poisson model has only one parameter denoted as
λ--> representing the mean number of events per unit of time or space The unit of time usually is chosen to be short enough that the mean arrival rate is not large λ<20 aka model of rare events
population variance
σ² the sum of the squared deviations from the mean divided by their population size
𝜋 is the binomial parameter for the probability of success on a single trial...what is it for sampling distributions
𝜋 is 𝑝 bar=𝑋 / n
only parameter needed to define a Bernoulli process
𝜋: mean 𝜋(1-𝜋): Variance